Skip to content

Synthetic Images

Jump to bottom
Hervé Bitteur edited this page May 29, 2017 · 6 revisions

MuseScore handles numerous scores under their symbolic format and can thus generate the information needed by the OMR dataset, based on MuseScore own layout:

  • Any symbol which is painted on the output score can lead to a symbol annotation containing its shape name and its bounding box.

  • A degree of variation is brought by the use of different musical fonts. MuseScore can select different fonts and thus produce outputs with slight variations.

For more details, see MuseScore organization on GitHub. The branch dealing with this generation software is on https://github.com/musescore/MuseScore/tree/imeta

ZHAW university uses a similar approach, starting from MuseScore data available in the community:

  • They run Lilypond software on MuseScore data, and thus use Lilypond own layout (which can be somewhat different from MuseScore layout). They then take advantage of the Lilypond MusicXML output which contains very precise symbol location information, and can thus be used to generate the symbols annotations.

Introduction

Dataset Model

Dataset Production

Training

To Do

Clone this wiki locally